AN EFFICIENT LIVER CANCER PREDICTION BY ADAPTIVE BP-NEURAL NETWORK AND BAT ALGORITHM

AN EFFICIENT LIVER CANCER PREDICTION BY ADAPTIVE BP-NEURAL NETWORK AND BAT ALGORITHM

1. Introduction

To study the largest internal organ of the human being that can be considered as the liver. It can be seen at the right ribs exactly beneath our lung right side. It can be regarded as various kinds of lobes. The primary cells of the liver can be considered as Hepatocytes. They also contain multiple types of cells in the liver, such as line its vessels and line small tubes in the large organ of the liver that can be considered as bile ducts. That can be used to carry the Bile in the organ through the gallbladder or otherwise known as directly to the intestine. The various kinds of liver cells that can contain many types of cancerous and non-cancerous. The mortal can be considered as malignant, and the non-cancerous can be regarded as benign tumours. The liver tumours have various causes, can be considered as multiple kindly, and different prognosis. Several varieties of cancer can shape in the liver. The most not unusual form of liver most diseases is hepatocellular carcinoma, which starts evolved inside the essential way of liver mobile (hepatocyte). Different styles of maximum liver cancers, consisting of intrahepatic cholangiocarcinoma and hepatoblastoma, are plenty much less not unusual. Pc Aided prognosis (CAD) device may be very beneficial for all the scientific experts in detection and diagnosing abnormalities earlier and faster. It’s miles a second opinion for the scientific specialists before suggesting tissue segmentation biopsy take a look at. Numerous strategies are available for liver maximum cancers prognosis, but those strategies are higher luxurious, time-ingesting, and having much less capability and accuracy for detecting the liver most cancers.

Consequently, a unique prediction method is essential to be waiting for the maximum lung cancers in its new ranges. In this paper, an integrated framework for predicting liver cancer is brought the usage of Adaptive B.P. Neural network with a B.A.T. set of rules. Neural networks offer one-of-a-kind trouble solving strategies wherein the neurons are being educated and examined with the aid of way of using the given database. The liver’s maximum cancers capabilities are extracted for predicting the cancer stage primarily based on the positive function used inside the system. Function selection is used to discover the subsets of maximum cancers cells interior a database and reduce the wide variety of most cancers cells furnished to the computation technique.

Figure: 1 schematic representation of the adaptive neural network

Figure 1 represents the overall process representation of the Adaptive neural network. Here Better performance can be achieved by discarding some features.

Adaptive neural network:

The interconnected adaptive neural network contains neurons with the intention of transmits the pattern of the electrical signal. An Adaptive Neural Network (ANN) is a bunch of learning neurons based on the biological neural networks (human brain). Typically, a neural network includes about one hundred billion neurons, and every neuron can be connected as much as the ten thousand other neurons adaptive neural networks are commonly presented as systems “neurons,” which are all constant and which exchange of the messages most of the distinct neurons. The connections have today’s numeric weights that can be adaptive primarily based definitely on the experience, developing neural internet adaptive to that of the inputs and able to the gaining knowledge of. The gain of the ANNs is that they’re regularly suitable to resolve issues that can be too complex to be solved via all the different predictable strategies, or it is very tough to discover the algorithmic answers. Adaptive neural networks (ANN) or connectionist tool is a computing device that is stimulated via the uncooked given information deliver however no longer always identical to that of the biological neural networks that compose of the animal brains. Such structures are hard to perform obligations through considering examples through generally without being programmed with some other challenge-unique guidelines. The processing of the artificial neural network is already we can feed some of the information related to that of all the methods, objects and other algorithms. So it can quickly identify and make classification regarding the purposes clearly and effectively. Instead, they routinely generate figuring out developments from the mastering cloth that they manner. An ANN is based on a group of the associated gadgets or nodes model of the neurons in an organic brain. Every connection, similar to the synapses in a natural mind, that could transmit a signal from one synthetic neuron to every other complement. An artificial neuron that gets a sign which could way it and then signal more adaptive neurons that may be related to it. In typical ANN implementations, the signal at a connection between the artificial neurons is a real wide variety of the values, and the output of every artificial neuron is computed using way of a few non-linear characteristics of the sum of its inputs. The connections among the artificial neurons are referred to as us the ‘edges.’ Adaptive neurons and edges usually are having a weight that would alter the studying proceeds of the protocols. The burden increases or decreases with the power of the sign at a specific connection. Adaptive neurons can also have get admission to inclusive of that the sign is ship if the blended signal passes that threshold. Normally, adaptive neurons are combination one into numerous layers. Unique layers may moreover carry out individual styles of the changes depends upon the source of their input. Indicators tour fastly from the first layer (i.e. the input layer) to the remaining segment (the output layer) and probably after the traversing the sheets at multiple times. The original purpose of the ANN approach changed into to clear up issues within the identical manner that a human mind might.

Figure: 2 General process flow of the Adaptive neural network

Figure 2 represents a flow sheet of the adaptive neural network process. It is somewhat a general flow sheeting process here. We need to elaborate on the basic process methods.

2. RELATED WORKS:

2.1 Neural network:

Fooladi et al. (2008) [17] proposed that the sensitivity to the induction of chromosomal harm through ionizing Gamma publicity is, on average, higher in breast maximum cancers patients than of normal wholesome controls. The gamma impact in every body’s lymphocytes and the evaluation between two organizations became tested, seventy-two hours after blood pattern culturing, using manner of revealing the samples to gamma rays after which they have been harvested. The exposure of gamma rays causes abnormality in chromosomes. The database used in this proposed method consists of chromosome breakage in seven chromosome corporations and the age of patients. on this technique, precept element evaluation (P.C.A.) is used for a symbolic selection degree. Then artificial Neural Networks (ANN) is used for form of ordinary times from extraordinary cases. The experimental assertion indicates that the result is obtained with an accuracy price of ninety 3.09% for Neural Networks (N.N.) classifier. Shukla et al. (2009) proposed a unique technique to simulate a knowledge-based completely gadget for the prognosis of breast maximum cancers, the usage of soft Computing gear like synthetic Neural Networks (ANNs) and Neuro-Fuzzy systems. The feed-ahead neural network has been skilled the use of three ANN algorithms especially again Propagation set of rules, Radial foundation function (RBF) Networks and the mastering Vector Quantization (L.V.Q.) Networks, Adaptive Neuro-Fuzzy Inference device (A.N.F.I.S.). The simulation become completed the usage of M.A.T.L.A.B. and overall performance have become evaluated using considering the metrics like the accuracy of analysis, education time, amount of neurons, range of epochs and lots of others., and those parameters may be very effective for early detection of breast most cancers. Steven et al.(1993) proposed about the synthetic neural networks have confirmed to be an interesting and useful change processing method. artificial neural techniques aren’t magical solutions with mystical talents that work without suitable engineering. With well statistics of their abilities and obstacles they may be achieved productively to issues in early detection and diagnosis of cancer. the proper most cancers packages so one can be used to demonstrate present day-day paintings in artificial neural networks for maximum cancers detection and diagnosis are breast cancer, liver most cancers and lung maximum cancers.[1] Wen Li et al. (2015) proposed an automatic technique based mostly on convolution neural networks (CNNs) is provided to section lesions from C.T. photographs. The CNNs is one in every of deep getting to know models with some convolution filters that could look at hierarchical capabilities from records. We, in comparison, the CNNs model to great gadget gaining knowledge of algorithms: AdaBoost, Random Forests (R.F.), and help vector system (SVM). those classifiers had been skilled through the use of handmade talents containing recommend, variance, contextual competencies [2] Philip et al(1994) proposed a options and developed a plan to improve the diagnostic accuracy of the networks. That plan consists of: (i) get many greater affected character instances and greater fact variables, which include M.R.I. and C.T. facts, so the network may be greater tremendously trained. a scarcity of sufficient affected person cases to properly train the community is the key trouble; (ii) use genetic algorithms and different techniques to preprocess the facts; (iii) the network want to have an optical interface to study pictures immediately; and (iv) build a patron-friendly interface using the interval on a 486 microcomputer. persisted research alongside the recommendations in this take a look at need to provide an advanced neural network for early detection of hepatic maximum cancers that hopefully will exceed the diagnostic talents of most radiologists.[3]Philip et al. (1992) proposed that approximately an again-propagation neural network changed into designed to diagnose five classifications of hepatic loads: Hepatoma, metastatic carcinoma, abscess, cavernous Hemangioma, and cirrhosis. The network input consisted of 35 numbers in step with affected person case that represented ultrasonographic records and laboratory checks [4] saribaet al. (2013) proposed about the general overall performance of the ANN and SVM classifiers on four fantastic most cancers datasets. For breast most cancers and liver maximum cancers dataset, the capabilities of the information are based on the situation of the organs  which is likewise referred to as widespread records even as for prostate cancer and ovarian cancer; each of those datasets are in the shape of gene expression facts. The datasets together with benign and malignant tumours is particular to classify with proposed techniques. The performance of every classifier is evaluated by the use of four awesome measuring devices that are accuracy, sensitivity, specificity and vicinity underneath Curve (A.U.C.)[5] ] Shika et al. (2015) proposed approximately the various neural community technology for the category of most cancers. the precept purpose of this survey in clinical diagnostics is to guide researchers to expand maximum fee effective and character satisfactory structures, techniques and strategies for clinicians.[6] Jae sung hong et al proposes about an automatic tool that could perform the entire diagnostic method from the extraction of the liver to the recognition of a tumor. Mainly, the proposed technique makes use of form records to emerge as aware of and apprehend a lesion adjacent to the border of the liver, that could in any other case be disregarded. Because of the fact such a place is concave like bay, morphological operations may be used to discover the bay. Further, for the reason that depth of a lesion can range appreciably consistent with the affected person and the slice taken, a preference on the threshold for extraction isn’t always clean. Therefore, the proposed approach extracts the lesion through using a Fuzzy c-manner clustering approach, which may determine the brink irrespective of a converting depth. furthermore, as a manner to lower any faulty diagnoses, the proposed device plays a 3-D consistency take a look at primarily based totally on 3-dimensional facts that a lesion mass can’t seem in a unmarried slice independently. [7]Shi et al 2016 proposed a overview of neural networks used in medical picture processing. We classify neural networks with the aid of the use of its processing dreams and the character of scientific pictures. Principal contributions, blessings, and disadvantages of the strategies are said within the paper. complicated problems of neural network application for clinical picture processing and an outlook for the destiny research are also discussed[8]Rauet al(2016)Proposed approximately the Diabetes mellitus that is related to an advanced threat of liver maximum cancers, and people two illnesses are a number of the most commonplace and vital causes of morbidity and mortality in Taiwan. to use statistics mining techniques to increase a model for predicting the improvement of liver most cancers inner 6 years of analysis with type II diabetes. techniques: information were received from the countrywide health insurance research Database[9]Gornesu et al. (2012) proposed about a synergetic gadget, primarily based on each precise statistical device and the sensitivity analysis furnished via neural networks is used for lowering the measurement of the database from twenty-5 to simply six attributes. An evolutionary-skilled neural network is superior afterwards for the class of the liver fibrosis degrees. The tandem approach is direct and easy, because of embedding the characteristic choice gadget into the approach form, if you want to dynamically pay attention the search handiest on the most applicable attributes. Experimental effects and an intensive statistical evaluation virtually verified the overall performance of the proposed sensible gadget in evaluation with distinct gadget studying techniques said in literature.[10] Poon et al. (2011) proposed about the category bushes and neural networks that diagnosed serological liver marker profiles comprising AFP, •1-antitrypsin (A1AT), •2-macroglobulin (A2MG), thyroxin-binding globulin (TBG), transferring and albumin similarly to sex and age, which might also allow the evaluation of H.C.C. [11] Adrian et al. (2012) proposed about the accuracy of real-time E.U.S. elastography in focal pancreatic lesions using pc-aided diagnosis thru synthetic neural community assessment.[12] Deepalakshmi et al. (2012) proposed about the technique of most vital factor evaluation is used to extract the most vital skills or hints of maximum information from the facts set there with the aid of robotically deciding on appropriate functions. The use of those greatest capabilities, a very last combined feature set, is formed and is hired for the type of the liver lesions into individual instructions. K-manner clustering and neural community based automated classifiers are employed on this system. The classifier design and its overall performance are studied and diverse statistical and spectral texture parameter extraction techniques, most appropriate characteristic desire strategies and automated class methods.[13] Deepti Mittal et al. (2011) proposed the pc-aided diagnostic system to assist radiologists in identifying focal liver lesions in B-mode ultrasound images. it can be used to discriminate focal liver sicknesses, which include Cyst, Hemangioma, Hepatocellular carcinoma and Metastases, at the aspect of normal liver.[14]Judith et al. (2001) proposed approximately the bin‐model” techniques, wherein affected character outcome and manipulated is thought from the statistical organizations in which the affected individual fits. The authors offer a cause of that with neural networks. It’s far feasible to mediate predictions for individual sufferers with incidence and misclassification price issues using receiver working specific technique. The authors illustrate their findings with examples that consist of prostate carcinoma detection, coronary heart sickness threat prediction, and remedy dosing. The authors emerge as aware of and talk about limitations to achievement, along with the need for increased databases and the need to set up multidisciplinary corporations. The authors consider that these boundaries can be conquered and that neural networks have a very critical feature in destiny clear choice help and the affected person manipulate systems hired in recurring medical exercise. [15] Amato et al. (2013) proposed the overall philosophy for the use of ANNs in diagnostic techniques through decided on examples, documenting the vast variability of records that could function inputs for ANNs. Interest will not only accept the strength of ANNs programs, but additionally to an evaluation in their limits, possible trends, and future inclinations and connections to other branches of human medicine [16]. Hsu et al. (2019) proposed the clean detection of the breast cancer technique through the usage of enforcing a unique method, after which compares the method with the logistic regression and SVM classifier. [18]

3. PROBLEM STATEMENT:

Liver maximum cancers are a malignant disorder with constrained useful options due to its competitive evolution. It locations heavy burden on maximum low and middle-income nations to treat the liver most cancers patients. Now, the accurate liver most cancers hazard predictions can help to make results on the need for the liver maximum cancer surveillance and beginning the better remedy for the patients. So we need to search for a powerful method for an appropriate prediction of the early liver most cancers.

4. PROPOSED METHODOLOGY:

Our work is based on the Bat Algorithm. We place an Adaptive B.P. neural network method for effective liver cancer prediction by improving the high percentage of accuracy.

Figure: 3 schematic illustrations of the proposed method

The above fig represents an entire process for the purpose of the effective liver cancer prediction

 Liver cancer data set:

The liver computer tomography data sets can be obtained from several sources, and it can be used for further other processing methods.

 Data preprocessing:

The initial step for liver cancer detection is the preprocessing step to fill up the missing data and to eliminate the unnecessary information from the dataset. In this process, we are using the watershed algorithm for the effective segmentation of the liver from the C.T. images. For that process;

1. First, we need to find the gradient of the picture.
2. Then after that applying the Watershed algorithm for the angle.
3. Then applying morphology to get the liver segment.
4. Then the cancer lesion can be segmented using the Gaussian model

 Gaussian mixture model:

Here there is a set of data points that will be grouped into several parts or clusters based on their similarity. In the machine learning method, this is known as the method of clustering. This method here we are used for the efficient isolation of the lesion areas in the liver

By applying this method,

1. We need to create the histogram
2. After that, we need to meet the expectation of the histogram
3. Then maximization of the result will occur
4. Then create a mask on the lesion area.
5. Then quickly, we will isolate the affected area.

In one dimension of the probability distribution model, the Gaussian mixture is given by,

G (=(1)

The multivariate of the Gaussian distribution is provided by

G (=exp (-(2)

The probability density for a K cluster is given by,

P(X) =

From Bayer’s theorem,

P(X) = (4)

We have to rearrange the terms concerning µ then we get,

(5)

Similarly by taking the derivate concerning µ, K

(7)

 Feature Extraction:

Liver cancer has different features, and it is significant to extract for minimizing the difficulty in the detection process. The tumor caused due to the multiplication of cancer cells (functions) in the lung, which is to be removed in this detection system. This feature extraction process is performed by adopting PSO. The feature extraction is the part of the pattern recognition techniques done on the input data for retrieving the relevant and collect the information about cancer to predict the patient conditions for further interpretations. The characteristic extraction begins off evolved from an initial set of measured facts and builds derived values imagined to be informative and facilitating the subsequent learning and generalization steps, and in a few instances leading to better human interpretations. Feature extraction is related to dimensionality reduction.

The enter facts to a set of rules are indeed too large to be processed, and it is suspected to be redundant, then it can be converted into a discounted set of abilities. Identifying a subset of the preliminary functions is called function desire. The chosen features are anticipated to comprise the relevant information from the enter data, just so the favoured mission can be executed through using this decreased illustration in the region of the whole preliminary facts.

Particle swarm optimization:

It is an evolutionary technique that is mainly used for the process of evolutionary programming, evolutionary methods, and genetic programming. PSO is sociologically inspired since the algorithm is based totally on sociological behaviour related to the approach of hen flocking. It is a population-based evolutionary algorithm that’s much like the different population primarily based evolutionary algorithms, PSO is used to attain answer a few of the random people. It is an appearance inside the form of the disorganized community of the shifting debris that generally tends to cluster collectively on an identical time as each particle can seems to be moving in an arbitrary course.

(8)

The position of each vector is enriched with a new vector in the form of,

(9)

PSO decreases linearly concerning optimization,

W=

Training and testing set:

The input data samples which are trained and tested by using the adaptive B.P.neural network. Initially, the weights of input data from the network are chosen arbitrarily. They are trained with a data sample for learning and to perform the process of the classification process, then with the testing dataset. The classification result of the tested data can be weighed to check the frequency error or the error rate, which occurs during the classification process, and the error can occur means we can solve the error by changing the weights of initial the data set.

 ABP Neural network with B.A.T. algorithm:

            Here in this paper, we describe a neural network with backpropagation. It’s going to work a long way faster than that of the sooner procedures to gaining knowledge of, makes it possible to apply the neural nets to solve troubles which had formerly been not available to solve. We are starting the process by the activation of the neuron,

(11)

The equation can be written  in vectorized form,

(12)

The goal of backpropagation is written in the quadratic form,

C= (13)

The C function is shown below,

C=(14)

The same C function is shown below,

(15)

The quadratic set in which the training set can be merged,

The error in the neuron can be represented as,

The equation can be represented in the form of the matrix,

=(

For back-propagating the error,

*

The gradient output is given by,

(20)

Steps for Neural Network:

Basic steps for developing a neural network

1: the usual preparation

2: initialization

3: forward propagation

4: backwards propagation

5: the training phase

Primary neural network utilizing a procedure

Input- Liver Cancer Dataset

Output- trained adaptive Neural Network

1- Gave an input

2 –Weighing the Input

3 – Sum of all the weighted data (average)

B.A.T. Algorithm:

Bat algorithm because of its acquaintance, illustrious global search performance and its compatibility even with the ample search space to find the best solutions just based on a performance measure. The functionality of the echolocation of microbats is thrilling as these bats can locate their prey and distinguish different varieties of insects even in whole darkness. We are able to first formulate the bat set of rules by using idealizing the echolocation behavior of bats. Our main aim is to integrate our adaptive B.P. neural network with the Bat algorithm based classifier.

Figure: 4 Process flow sheet for the B.A.T. algorithm

Figure 4 represents the overall methodology, followed by the B.A.T. algorithm.

A virtual Bat movement is represented by,

(22)

Pseudo code for the mutated agorithm

Objective function f(x), x = (x1, …, xd) T

Initialize the bat population xi (i = 1, 2, …, n) and vi

Define pulse frequency (fi at xi)

Initialize pulse rates (ri) and the loudness (Bi)

while (t>Maximum of iterations)

Select a solution among the best solutions

Generate a local solution around the selected best solution

end if

Generate a new solution by flying randomly

 if (rand < Ai & f(xi) < f(x∗))

Accept the new solutions Increase (ri) and reduce (Bi)

end if

Rank the bats and find the current best (x∗)

end while

Post process results and visualization

Disease Diagnosis:

Finally, the disease can be diagnosed and then the clinical experts can read the results and predict whether it’s a cancerous or non-cancerous one.

5. RESULTS:

The data sets of the liver cancer C.T. images were obtained from the different patient

       1                         2                    3                 4                      5                    6    

           7                    8                   9                  10                    11                   12   

Fig: 5 Data sets of different cancer patients

Figure 5 represents the different mixed images of healthy and cancer patients.

Preprocessing:

Figure:6 step of preprocessing

Figure 6 represents that in the preprocessing step, the liver area could get bolded. So it is straightforward to segment in the next level of segmentation.

Segmentation:

Figure: 7 segmented forms of the liver area

Figure 7 represents that by using the Gaussian mixture model, we can easily segment the liver area.

Localization and classification using the proposed Adaptive B.P. neural network and B.A.T. algorithm:

Figure:8  localization of the cancer division in the liver

Figure 8 represents the accurate localization of cancer in the part of the liver.

Classification:

    Slice1(Benign)             Slice2(Benign)           Slice 3 (Malignant)    Slice4 (Malignant)

Figure:9  Adaptive Neural B.P. classification of the stages of cancer     

Figure 9 represents that the A.N.B.P. method of classification is accurate, and it can classify the benign and malignant stages of cancer.

The performance evaluation of proposed adaptive B.P. Neural Networks with B.A.T. Optimization is simulated. The implementation of this analysis is performed on the liver cancer dataset. The liver dataset was given as input to the network, and the information is divided into testing facts and check records. The training set for the neural network consists of 70% of the whole dataset, and the checking outset is 30% of the entire facts.

Accuracy

It is a narrative of the organized errors, a measure of arithmetical predisposition, low accuracy causes a variation between a consequence and a “true” value. ISO calls this actuality. This means that many times the unusual samples or images are tested with the same algorithm and the machine or system provides results on how much precise. The proposition of the actual results in the entire area it should be negative or positive.

                             Accuracy (A) = (TP+TN)/(TP + TN + FP + FN)

Precision:

Precision is a portrayal of random errors that is a measure of algebraic variability.

                                                Precision= TP/ (TP+ FP)

Sensitivity:

Sensitivity, moreover known as the exact constructive charge, the endure in thoughts, or opportunity of detection in a few fields measure the proportion of actual positives which can be successfully recognized.

                                                Sensitivity=TP/ (TP+FN)

Specificity:

Specificity, also called the true pessimistic range, determines the amount of the actual negatives that are defined incorrect manner.

                                                   Specificity) = TP/(TP+FP)

Table:1 Accuracy comparison of the proposed A.B.P.N.N-BAT algorithm

Figure: 10 Accuracy comparison of the proposed A.B.P.N.N-BAT algorithm

The above table and figure 10 represents the accuracy comparison of the proposed A.B.P.N.N-BAT algorithm. These attain the Classification accuracy of 97%, and it accurately predicts the Liver cancer by using the proposed method.

Table: 2 Specificity comparison of the proposed A.B.P.N.N-BAT algorithm

Figure: 11 Specificity comparison of the proposed A.B.P.N.N-BAT algorithm

Table 2 and figure 11 represent that shows that the specificity comparison for the proposed A.B.P.N.N-BAT algorithm. The Classification specificity of 95% and it accurately predicts the Liver cancer can be occurred by the proposed method.

                   Table: 3 Sensitivity comparison of the proposed A.B.P.N.N-BAT algorithm

Figure: 12 Sensitivity comparison of the proposed A.B.P.N.N-BAT algorithm

Table 3 and figure 12 shows that the sensitivity comparison for proposed E.B.A. classifier. The Classification sensitivity of 96% and it accurately predicts the Liver cancer can be occurred by the proposed method.

Table: 4 Precision comparison of the proposed A.B.P.N.N-BAT algorithm

Figure: 13 Precision comparison of the proposed A.B.P.N.N-BAT algorithm

Table 4 and figure 13 shows that the Precision comparison for proposed A.B.P.N.N-BAT algorithm. Classification precision of 94% and it accurately predicts the Liver cancer.

Therefore, the proposed method achieves effective results of the proposed parameter are compared to other existing classifiers.

6. CONCLUSION:

Detection of liver cancer is the most challenging problem in the medical field due to the configuration of the cancer cell, where most of the cells are overlapped fatty and collide with each other. There are over different types of cancer and one of them. The treatment of cancer that occurs in the liver should be delay results that can lead them to a high risk of the transience rate. Detection of liver cancer in the earlier stage it might be curable. In this work, liver cancer detection based on the adaptive B.P. neural network and B.A.T. algorithm get implemented. ANN has many of the advantages such as long training time, high computational cost, and adjustment of mass. They are mainly focused on providing efficient as well as early detection of liver cancer, and it’s also a cost and time-saving tactic. The performance evaluation of the proposed method shows effective results, and it indicates that an adaptive B.P. neural network should be can be protectively to detect liver cancer. The prediction could help medical experts to execute a proper medication and require the quick response of the patients.

7. References:

[1] Rogers, S. K., Ruck, D. W., & Kabrisky, M. Artificial neural networks for early detection and diagnosis of cancer. Cancer Letters, 77(2-3), (1994): 79–83

[2] Wen Li1,2, Fucang Jia1, Qinmao Hu1Automatic Segmentation of Liver Tumors in C.T. Images with Deep Convolutional Neural Networks, Journal of Computer and Communications, 2015

[3] Maclin, P. S., & Dempsey, J. How to improve a neural network for early detection of hepatic cancer. Cancer Letters, 77(2-3), (1994). 95–101

[4] Maclin, P. S., & Dempsey, J. Using an artificial neural network to diagnose hepatic masses. Journal of Medical Systems, 16(5), 215–225. (1992).

[5] Sy Ahmad Ubaidillah, S. H., Sallehuddin, R., & Ali, N. A. Cancer Detection Using Artificial Neural Network and Support Vector Machine: A Comparative Study. Jurnal Teknologi, (2013):65(1).

[6] Agrawal, S., & Agrawal, J. Neural Network Techniques for Cancer Prediction: A Survey. Procedia Computer Science, 60, (2015):769–774.

[7] Jae-sung hong  toyohisa Kaneko  ryuzo sekiguchi  kil-houm park Automatic liver cancer detection from ictieice transactions on information and systems   vol.e84-d   no.6 (2001): pp.741-748

[8] Zhenghao ,ShiLifeng He, Kenji Suzuki, Tsuyoshi Nakamura, and Hidenori Itoh Survey on Neural Networks Used for Medical Image Processing Int J Comput Sci. February ; 3(1): 2009:86–100.

[9] Rau, H.-H., Hsu, C.-Y., Lin, Y.-A., Atique, S., Fuad, A., Wei, L.-M., & Hsu, M.-H. Development of a web-based liver cancer prediction model for type II diabetes patients by using an artificial neural network. Computer Methods and Programs in Biomedicine, 125 (2016):58–65.

[10] Gorunescu, F., Belgium, S., Gorunescu, M., & Badea, R. Intelligent decision-making for liver fibrosis stabilization based on tandem feature selection and evolutionary-driven neural network. Expert Systems with Applications, 39(17), (2012): 12824–12832.

[11] Poon, T. C.-W., Chan, A. T.-C., Zee, B., Ho, S. K.-W., Mok, T. S.-K., Leung, T. W.-T., & Johnson, P. J. Application of Classification Tree and Neural Network Algorithms to the Identification of Serological Liver Marker Profiles for the Diagnosis of Hepatocellular Carcinoma. Oncology, 61(4), (2001). 275–283

[12] Săftoiu, A., Vilmann, P., Gorunescu, F., Janssen, J., Hocke, M., Larsen, M., … Ciurea, T. Efficacy of an Artificial Neural Network–Based Approach to Endoscopic Ultrasound Elastography in Diagnosis of Focal Pancreatic Masses. Clinical Gastroenterology and Hepatology, 10(1), (2012).  84–90

[13] Balasubramanian, D., Srinivasan, P., & Gurupatham, R. Automatic Classification of Focal Lesions in Ultrasound Liver Images using Principal Component Analysis and Neural Networks. 29th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. (2007).

[14] Mittal, D., Kumar, V., Saxena, S. C., Khandelwal, N., & Kalra, N. Neural network-based focal liver lesion diagnosis using ultrasound images. Computerized Medical Imaging and Graphics, 35(4), 315–323. (2011).

[15] Judith E. Dayhoff, Artificial neural network, Wiley online library,2011

[16] Amato, F., López, A., Peña-Méndez, E. M., Vaňhara, P., Hampl, A., & Havel, J. Artificial neural networks in medical diagnosis. Journal of Applied Biomedicine, 11(2), (2013). 47–58.

[17] Fooladi, A. A. I., Sattari, M., Hassan, Z. M., Mahdavi, M., Azizi, T., & Horii, In vivo induction of necrosis in mice fibrosarcoma via intravenous injection of type B staphylococcal enterotoxin. Biotechnology Letters, 30(12), (2008), 2053–2059

[18] Hsu, S.-M., Kuo, W.-H., Kuo, F.-C., & Liao, Y.-Y. Breast tumour classification using different features of quantitative ultrasound parametric images. International Journal of Computer Assisted Radiology and Surgery. doi:10.1007/s11548-018-01908-8 (2019).